Harnessing multimodal AI to reduce virologic failure in people living with HIV - artificial intelligence (AI) shows significant promise in optimizing HIV prevention and treatment efforts by…

The National Institute of Mental Health awarded San Diego State University Foundation $1.137 million in a Mental Health Research Grant (CFDA 93.242) on August 1, 2026, to develop and validate multimodal artificial intelligence models predicting virologic failure among persons living with HIV.

The award funds a two-year longitudinal cohort study of 200 patients at a large U.S.-based HIV clinic that integrates data from the Behavioral Engagement and Adherence Monitoring (BEAM) mobile health system with wearables and other multimodal indicators to identify modifiable factors associated with HIV treatment failure. Work includes developing knowledge graphs informed by focus groups of persons living with HIV and scientific literature to map interconnections among virologic failure predictors; designing BEAM for real-world implementation; and applying knowledge graphs and AI models to forecast virologic failure while minimizing algorithmic bias. A Community Advisory Board of persons living with HIV provides iterative feedback across all study phases to ensure models conform to AI ethical principles and remain patient-centered. Performance occurs in San Diego, California, with a completion date of April 30, 2031.

Harnessing multimodal AI to reduce virologic failure in people living with HIV - artificial intelligence (AI) shows significant promise in optimizing HIV prevention and treatment efforts by identifying modifiable behavioral or environmental factors and enhancing intervention strategies to deliver more impactful interventions. This proposal focuses on identifying factors that predict virologic failure among persons living with HIV (PLWH) in the united states (US). We propose to design and integrate multimodal data from a mobile health system called behavioral engagement and adherence monitoring (BEAM), guided by human- centered design and AI ethical principles, to capture deeper contextual information surrounding HIV virologic failure in a two-year longitudinal cohort study of PLWH at a large US-based HIV clinic. Multimodal data will be collected to develop accurate, safe, efficient, and unbiased AI models that, in conjunction with knowledge graphs (KGS), will predict virologic failure among PLWH. Throughout this process, a community advisory board (CAB) of PLWH will provide iterative feedback on all aspects of the study to ensure the models conform to AI ethical principles, minimizes bias, and are patient centered. We propose the following aims: aim 1: with extensive input from the CAB, focus groups, and scientific literature, we will identify key predictors of HIV virologic failure, and their interconnections, and develop a knowledge graph to inform the design and implementation of BEAM (behavioral engagement and adherence monitoring). We will conduct four focus groups (N=8/group) of PLWH and integrate findings with scientific evidence to develop initial drafts of KGS. Aim 2: implement BEAM in a 24-month longitudinal cohort of 200 patients at a large US-based HIV clinic to capture multimodal indicators of virologic failure risk allowing for real-world validation and adaptation of the knowledge graphs. We will collect participant data from wearables (i.e., fitbits), medication event monitoring systems (MEMS), surveys, ecological momentary assessment (EMA) surveys, and a mobile APP over a 12-month period, and clinical data from electronic health records (EHR) over a 24-month period. Aim 3: using a human-centered and ethical approach, develop AI models to predict HIV virologic failure and iteratively refine knowledge graphs. AI models will be constructed through pre-processing the data, model training and evaluation, and integrating knowledge graphs. Aim 4: leverage AI models and knowledge graphs to identify and CO-develop, with the CAB, intervention use cases to deliver tailored behavioral supports to at-risk PLWH and evaluate them through theater testing. We will CO-create with the CAB novel intervention strategies by leveraging AI-predicted risk factors and knowledge graph-derived insights and conduct theater testing with PLWH (N=8) and key stakeholders (N=8). The proposal is significant because rate of viral suppression among PLWH in the US is only at 65%, despite decades of efforts to engage PLWH in care. The public health impact of this proposal is strengthened by the application of the model to multiple lines of promising interventions to maximize their impact, to be tested more fully in future trials.